4 files changed, 59 insertions, 9 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index 5e7e5d2ff9c..bb4f5641616 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -122,7 +122,9 @@ static const char *semantic_names[] =
    "GENERIC",
    "NORMAL",
    "FACE",
-   "EDGEFLAG"
+   "EDGEFLAG",
+   "VERTICES_IN",
+   "PRIM_ID"
 };
 
 static const char *immediate_type_names[] =
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 123117cb0a3..e2f28783cc2 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -372,6 +372,7 @@ tgsi_exec_machine_create( void )
    memset(mach, 0, sizeof(*mach));
 
    mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR];
+   mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES;
    mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0];
 
    /* Setup constants. */
@@ -1468,6 +1469,15 @@ store_dest(
       index = mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
          + reg->Register.Index;
       dst = &mach->Outputs[offset + index].xyzw[chan_index];
+#if 0
+      if (TGSI_PROCESSOR_GEOMETRY == mach->Processor) {
+         fprintf(stderr, "STORING OUT[%d] mask(%d), = (", index, execmask);
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (execmask & (1 << i))
+               fprintf(stderr, "%f, ", chan->f[i]);
+         fprintf(stderr, ")\n");
+      }
+#endif
       break;
 
    case TGSI_FILE_TEMPORARY:
@@ -1638,6 +1648,35 @@ exec_kilp(struct tgsi_exec_machine *mach,
    mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
 }
 
+static void
+emit_vertex(struct tgsi_exec_machine *mach)
+{
+   /* FIXME: check for exec mask correctly
+   unsigned i;
+   for (i = 0; i < QUAD_SIZE; ++i) {
+         if ((mach->ExecMask & (1 << i)))
+   */
+   if (mach->ExecMask) {
+      mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += mach->NumOutputs;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
+   }
+}
+
+static void
+emit_primitive(struct tgsi_exec_machine *mach)
+{
+   unsigned *prim_count = &mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0];
+   /* FIXME: check for exec mask correctly
+   unsigned i;
+   for (i = 0; i < QUAD_SIZE; ++i) {
+         if ((mach->ExecMask & (1 << i)))
+   */
+   if (mach->ExecMask) {
+      ++(*prim_count);
+      debug_assert((*prim_count * mach->NumOutputs) < mach->MaxGeometryShaderOutputs);
+      mach->Primitives[*prim_count] = 0;
+   }
+}
 
 /*
  * Fetch a four texture samples using STR texture coordinates.
@@ -3087,13 +3126,11 @@ exec_instruction(
       break;
 
    case TGSI_OPCODE_EMIT:
-      mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += 16;
-      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
+      emit_vertex(mach);
       break;
 
    case TGSI_OPCODE_ENDPRIM:
-      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]++;
-      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0;
+      emit_primitive(mach);
       break;
 
    case TGSI_OPCODE_BGNFOR:
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index fd94c1bc440..afaf5c39c48 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -191,6 +191,14 @@ struct tgsi_exec_labels
  */
 #define TGSI_EXEC_MAX_CONST_BUFFER  4096
 
+/* The maximum number of vertices per primitive */
+#define TGSI_MAX_PRIM_VERTICES 6
+
+/* The maximum number of primitives to be generated */
+#define TGSI_MAX_PRIMITIVES 64
+
+/* The maximum total number of vertices */
+#define TGSI_MAX_TOTAL_VERTICES (TGSI_MAX_PRIM_VERTICES * TGSI_MAX_PRIMITIVES * PIPE_MAX_ATTRIBS)
 
 /** function call/activation record */
 struct tgsi_call_record
@@ -201,7 +209,6 @@ struct tgsi_call_record
    uint ReturnAddr;
 };
 
-
 /**
  * Run-time virtual machine state for executing TGSI shader.
  */
@@ -214,8 +221,8 @@ struct tgsi_exec_machine
 
    float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
 
-   struct tgsi_exec_vector       Inputs[PIPE_MAX_ATTRIBS];
-   struct tgsi_exec_vector       Outputs[PIPE_MAX_ATTRIBS];
+   struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
+   struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
 
    struct tgsi_exec_vector       *Addrs;
    struct tgsi_exec_vector       *Predicates;
@@ -229,6 +236,8 @@ struct tgsi_exec_machine
 
    /* GEOMETRY processor only. */
    unsigned                      *Primitives;
+   unsigned                       NumOutputs;
+   unsigned                       MaxGeometryShaderOutputs;
 
    /* FRAGMENT processor only. */
    const struct tgsi_interp_coef *InterpCoefs;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c
index f000958bfc0..84e863b2375 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_text.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.c
@@ -791,7 +791,9 @@ static const char *semantic_names[TGSI_SEMANTIC_COUNT] =
    "PSIZE",
    "GENERIC",
    "NORMAL",
-   "FACE"
+   "FACE",
+   "VERTICES_IN",
+   "PRIM_ID"
 };
 
 static const char *interpolate_names[TGSI_INTERPOLATE_COUNT] =